Source: Brookings Institution

AI systems are often described using anthropomorphic language that portrays outputs as the system’s own decisions rather than the result of human design choices about data, objectives, and deployment. Research demonstrates that users consistently attribute human qualities to machines, such as empathy and intentionality, even though models like large language models generate responses statistically and lack any intent, consciousness, or true understanding. Design features like realistic voices and named personas causally increase users’ perception of human-likeness and trust, which may lead to oversharing, emotional attachment, and increased reliance.

The use of human-centered verbs such as “decides,” “sees,” or “recognizes” in policy discourse can impede clear regulatory implementation by failing to describe measurable performance parameters or failure modes. This framing risks obscuring the roles of developers, deployers, and operators in shaping AI outputs. It can also reinforce public misconceptions regarding AI autonomy and intelligence, while complicating accountability structures. More precise, operational vocabulary that clearly characterizes AI capabilities and design limitations could support transparency and governance.